AI 용어집
인공지능 완전 사전
Instrumental Variable
Variable used in causal inference to isolate the effect of a treatment on an outcome, being correlated with the treatment variable but independent of the error term.
Two-Stage Least Squares Estimator
Two-step estimation method where the first step predicts the treatment variable with the instrument, and the second step uses this prediction to estimate the causal effect.
Weak Instrumental Variable
Instrument showing weak correlation with the treatment variable, leading to biased estimators and unreliable statistical tests.
Overidentification Test
Statistical test verifying the validity of exclusion assumptions when multiple instruments are available by testing their consistency.
Wald Ratio
Simple instrumental estimator calculated as the ratio between the effect of the instrument on the outcome and its effect on the treatment.
Local Average Treatment Effect
Average causal effect identified by an instrumental variable, applying specifically to individuals whose treatment is modified by the instrument (compliers).
Complier
Individual whose treatment status follows the exogenous variation induced by the instrument, constituting the focal group of the LATE effect.
First-stage
First stage of 2SLS estimation where the treatment variable is regressed on the instruments to isolate its exogenous variation.
Reduced form
Direct regression of the outcome on the instrument, capturing the total effect of the instrument without specifying the intermediate causal mechanism.
Endogeneity
Problem where the explanatory variable is correlated with the error term, requiring the use of instrumental variables for causal identification.
Valid instrument
Instrumental variable simultaneously satisfying the relevance, exclusion, and independence assumptions for correct causal identification.
Monotonicity assumption
Condition stipulating that the instrument cannot push any individual to reduce their treatment, having only null or positive effects.
Never-taker
Individual who never takes the treatment regardless of the instrument's value, contributing to response heterogeneity.
Always-taker
Individual who systematically adopts the treatment regardless of the instrument's value, unaffected by instrumental variation.
Defier
Individual whose treatment status varies inversely with the instrument, potentially violating the monotonicity assumption.
First-stage F-statistic
Indicator of the strength of correlation between instruments and endogenous variable, with a value greater than 10 suggesting robust instruments.
Simultaneity bias
Endogeneity problem where the treatment and the outcome are mutually determined, justifying the instrumental variables approach.
Exogenous instrument
Instrumental variable statistically independent of the unobserved factors affecting the outcome, an essential condition for causal identification.